“ASR Status Report 2016-10-31”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
 
(相同用户的一个中间修订版本未显示)
第21行: 第21行:
 
|Ying Shi   
 
|Ying Shi   
 
||  
 
||  
*  DNN and TDNN visualization;CNN baseline(th30) new model DNN+lstm(th30);ML-book  
+
*  DNN and TDNN visualization;CNN baseline(th30) new model DNN+lstm(th30);ML-book [http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=shiying&step=view_request&cvssid=569 slide]
 
||  
 
||  
 
*  ML-book continue DNN visualization maybe I should find a new tool t-sne is not powerful enough
 
*  ML-book continue DNN visualization maybe I should find a new tool t-sne is not powerful enough

2016年11月2日 (三) 01:28的最后版本

Date People Last Week This Week


2016.10.31 Hang Luo
  • Know the source code of joint training.
  • Experiments on joint training, change the receive computation blocks and try non-symmetric structure.The result expected on this Thursday.
  • Find paper about joint training to read.
  • Try new experiments according to last two experiments.
Ying Shi
  • DNN and TDNN visualization;CNN baseline(th30) new model DNN+lstm(th30);ML-book slide
  • ML-book continue DNN visualization maybe I should find a new tool t-sne is not powerful enough
Yixiang Chen
  • Joint-training (GMM baseline)
  • Joint-training
Lantian Li
  • Joint-training on TASLP (DNN i-vector parts). [1]
  • Joint-training on SRE and LRE (Neural model is over-fitting...). [2]
  • Joint-training on TASLP.
  • Joint-training on SRE and LRE.
Zhiyuan Tang
  • O-COCOSDA 16, Bali.
  • report for Weekly Reading;
  • finish the additinal experiments of joint learning (speech & spk) for taslp (multi-task, ivector as part of input)[3].